It's more like a recipe (for functions).

The first example, I, is an identity function. It takes y and returns y.

The second, K, is a constant which takes X and y and returns x.

This gets more complicated as you go along. The idea is that you get rid of a lot of the syntax for composition and have it all be implicit by what you put next to each other (given APL programs are usually one long line of a bunch of different symbols all representing functions).

To complement leethomps answer, combinatory logic is a branch of Mathematics that was started in the 1920s by a mathematician called Moses Shönfinkel which deals with "functions that do stuff and return other functions".

This was developed by some names that may be more familiar (Haskell Curry, Alan Turing, Kurt Gödel, Bertrand Russell). It was proved to be identical to both the lambda calculus and the Turing machine and became the basis for modern computing.

What we see here are some of those key building blocks that were studied in the 20s and 30s and have been now applied to modern programming languages.

Functional languages use them a lot because you can express a lot of things as just combinations and compositions of other functions. Array languages often take this to an extreme by expressing complex numeric algorithms with only a few symbols.

What you see above is the logic/processing order of how those functions fit together. For example you can express a mean as something like `(+/#)` - a 5 letter anonymous function that can be applied to an array - because of all the applications and combinations being implicit in the structure of the language, as denoted in the link.